Automatic Extraction of DH Parameters of Serial Manipulators using Line Geometry

Transcription

1 The 2 nd Joint International Conference on Multibody System Dynamics May 29-June 1, 2012, Stuttgart, Germany Automatic Extraction of DH Parameters of Serial Manipulators using Line Geometry Rajeevlochana C.G.*, Subir K. Saha*, Shivesh Kumar # *Department of Mechanical Engineering Indian Institute of Technology Delhi New Delhi, , India rajeevlochan.iitd@gmail.com, saha@mech.iitd.ac.in # Department of Mechanical Engineering National Institute of Technology Karnataka Suratkal, , India shivesh.mecheng@gmail.com ABSTRACT Serial manipulators/robots are used extensively in industries to perform various tasks such as pick-andplace operation, painting, arc-welding, assembly of components etc. To perform tasks accurately, exact kinematic parameters of the manipulator are required. Note that these parameters are generally represented using the well-known Denavit-Hartenberg (DH) parameters. Typically, a set of nominal DH parameters are provided by the robot manufacturers, which may not be exact due to assembly errors etc. Hence, there is a need to know them exactly. In this paper, a novel analytic method is proposed to extract DH parameters of a robot manipulator. For this, each joint axis of a manipulator consisting of a direction and a point on it are provided as input to the proposed algorithm. The exact DH parameters are then extracted recursively from base link to the end-effector using the concepts of Plücker coordinates and Dual Vector Algebra. The algorithm has been implemented as an addin/plugin inside Autdoesk Inventor CAD software, which determines the DH parameters of a serial manipulator from its CAD model. 1. INTRODUCTION Serial manipulators/robots usually have a poor accuracy as compared to their repeatability. The reason could be a non-exact kinematic model, due to which a manipulator reaches configuration A instead of the desired configuration B. Since the repeatability is higher, the manipulator reaches configurations closer to A whenever it is asked to reach configuration B. The inaccuracy in kinematic model could be due to manufacturing errors, assembly issues, wear and tear, permanent bending of the links due to fatigue, etc. Note here that the well-known Denavit-Hartenberg (DH) parameters [4] are typically used to define the robot kinematics. Over the years, researchers have proposed various methods to determine these DH parameters. For example, [1] and [2] identified them directly, whereas [11] conceptualized and identified what the authors call S-Model parameters from which DH parameters are determined. Since the DH parameters are the most common and standard way of representing a robot architecture, some of the important methods to find them are described here. In [2], it requires the location and orientation of the joint axes as inputs, which are determined by rotating one joint at a time and locking the others. Such rotation will result in a circular motion by the end-effector. Hence, by tracing the endeffector positions one can determine the circle and its axis which is the axis of the revolute joint causing the circular motion. These joint axes are then used to determine the exact DH parameters of the serial manipulator at hand using Vector Algebra. In [1], the DH parameters were determined by applying the methodology introduced in [11], which uses the plane of rotation and the centre of rotation. Further it is extended to use the radius of rotation and a plane translated along the axis of rotation. The idea developed by [6] was also used in [1] where an extra parameter () was introduced to deal with parallel or near parallel joint axes. In this paper, we propose a novel analytical methodology to extract the DH parameters of a serial manipulator with revolute joints only using line geometry, Plücker coordinates and Dual Vector Algebra. It requires a point on the joint axes and its direction. It then determines whether two consecutive joint

2 axes are parallel, intersecting or skewed [7]. It then analytically determines the DH parameters using the Plücker coordinates [3] and the Vector Algebra. Implementation of the proposed methodology is simpler and elegant compared to the one reported in [2]. It determines one set of DH parameters at a time in a recursive manner from the base link to the end-effector, which is more efficient and elegant than determination using Paul s backwards multiplication technique [1]. The paper is organized as follows: Section 2 introduces the concept of line geometry, followed by Plücker coordinates and Dual Vector Algebra. Section 3 then gives the proposed analytical methodology. An implementation of the proposed methodology is presented in Section 4, followed by conclusions in Section LINE GEOMETRY A straight line in the 3D space can be represented using a unit vector (e) to represent its direction and a position vector (c) to represent a point on the line, both defined in a specific coordinate frame of reference (say, Frame F), as shown in Figure 1(a). Alternatively, Plücker coordinates [3] can be used to represent the same line using its unit directional vector (e) and the moment vector (k), as shown in Figure 1(b). Equation (1) defines the Plücker coordinates, where s is the 6-dimensional column vector. C C c e c k e (a) Line in 3D space (b) Plücker coordinates representing a line (c) Two lines in 3D space Figure 1: Line geometry in 3D space = = (1) In Equation (1), e is the 3-dimensional unit vector parallel to the line, and c is the 3-dimensional position vector of the point C on the line. The 3-dimensional vector k =, then corresponds to the moment vector Pair of Straight Lines Two lines in the 3D space are intersecting, parallel, collinear or skewed to each other. An elegant method, as reported in [7], to determine their relationship using the concepts of Dual Vector Algebra [5] is used here. Note that the Plücker coordinates used to represent a line can be expressed as a dual vector (), which has a real part () and dual part (), i.e., where ε 2 = 0. = +ε (2) For any two lines represented as dual vectors, say and, line dot product, as illustrated in Figure 1(c), results in a dual number whose dual part indicates if the lines are skewed or not. For non-skewed lines, the real part determines if the lines are parallel or intersecting. Consider this result of the line dot product [7] as given below:. = ( +ε ). ( + ε ) =. + ε(. +. ) (3)

3 = cos() +ε[ d sin()] where is the angle between the two lines, and d is the distance measured along common normal. In our proposed methodology, the above concept of line dot product is used to determine the relationship between two successive joint axes of the serial manipulator at hand Intersecting Lines Plücker coordinates of two lines s 1 and s 2 are used here to find their intersection point (P), as shown in Figure 2(a). The intersection point denoted with a vector p as reported in [3] is as below: =. or. for.. = 0 (4) P P P (a) Intersecting lines (b) One of the intersecting lines (c) Parallel lines passing through origin Figure 2: Intersecting and parallel lines The expression for vector p is more elegant and compact compared to the one using Line Geometry or Vector Algebra. If either of the line passes through the origin O F the moment k of that line becomes null (0), and hence the intersection point P cannot be determined using Equation 4. Hence, a novel method is proposed to overcome this inability by following the steps below. This is illustrated in Figure 2(b). 1. Find a point O G at unit distance from O F along the common normal of and, i.e., [ ] = [ ] [ ] (5) 2. Define a new coordinate frame (Frame G) with origin at O G, and the coordinate axes parallel to those in Frame F. The transformation from Frame F to Frame G is a pure translation with no rotation. 3. Vectors and denoting the directions are invariant after transformation from Frame F to Frame G, i.e., [ ] = [ ] ; [ ] = [ ] (6) 4. Vectors and are transformed to Frame G i.e., [ ] = [ ] [ ] ; [ ] = [ ] [ ] (7) 5. Vectors and being the moments are transformed to Frame G [5], i.e.,

4 [ ] = [ ] + [ ] [ ] ; [ ] = [ ] + [ ] [ ] (8) 6. Intersection point [] is then found in Frame G as [] = [] [ ] [ ]. [ ] or [] [ ] [ ]. [ ] (9) 7. Intersection point [] is next found in Frame F as [] = [ ] + [] (10) The above approach is used to find the intersection point of two successive joint axes of a serial manipulator, which plays an important role in the extraction of the DH parameters Parallel Lines Between two parallel lines ( and ), as shown in Figure 2(c), no unique solution exists for a line that is a common normal. One of the solutions can be determined by putting a condition that it passes through an arbitrary point C 1. To achieve this, a vector ( ) normal to and is determined as given below: = (11) where vectors and are indicated in Figure 2(c). Moreover, a unit vector ( ) normal to and is determined as follows: = (12) Next, a new set of Plücker coordinates defining a new line perpendicular to is introduced below: = (13) Using the lines and one can now find the point of intersection P or vector p using the methodology presented in Section Skewed Lines Between two skewed lines ( and ), a unique solution exists that is normal to both the lines, as shown in Figure 1(c). These points P 1 and P 2 on lines and respectively are determined using the Plücker coordinates [3] i.e., =. cos(). sin() + (14) =.+cos(). sin() + (15) where = is the common normal and is the angle between the lines as indicated in Figure 1(c). Both the vectors and are used to extract the DH parameters for two successive skewed joint axes.

5 3. DENAVIT-HARTENBERG (DH) PARAMETERS The links of a serial manipulator usually coupled with single-degree-of-freedom revolute or prismatic joints. A coordinate frame is attached to each link, as shown in Figure 3(a). The transformation between the Frame i+1 attached to Link i and the Frame i attached to Link i-1 can be represented using four Denavit-Hartenberg(DH) parameters [4], where i is the index of the link. Due to the space limitation, only the definitions of the parameters illustrated in Figure 3(b) are given below, whereas the details about the rules of attaching frames are available in [10]: a. Joint Offset (b i ): Distance between X i and X i+1 along Z i b. Joint Angle (θ i ): Angle between X i and X i+1 about Z i c. Link Length (a i ): Distance between Z i and Z i+1 along X i+1 d. Twist Angle (α i ): Angle between Z i and Z i+1 about X i+1 Frame 4 Frame 3 Link 3 Link 2 Link 1 Link 0 Frame 2 Frame 1 (a) Coordinate frames attached to each link (b) Two successive coordinate frames Figure 3: Representation of DH parameters 3.1. Extraction of DH Parameters The proposed analytical method to extract the DH parameters of an n-degree-of-freedom serial manipulator is presented here. For the sake of simplicity in calculations, the frame attached to the base link (Link 0) is identical to the fixed coordinate frame (Frame F) such that points O 1 is coincident with O F, and the first joint axis (Z 1 ) is along the Z F. For all the subsequent joints, the joint axes, consisting of unit vector ( ) and the vector representing a point on the axis need to be provided as input with respect to Frame F. The DH parameters are then extracted recursively from base link (Link 0) to the end-effector (Link n). For the first step, i.e., when i = 1, the unit vectors and corresponding to first two joint axes are known as they are provided as input. The origin ( ) of Frame 1 is also known from the assumption and a vector to a point ( ) on second axis is known as input. For each subsequent step i, the two joint axes are represented as Plücker coordinates ( and ) as given below: = [ ] T ; = [ ] T (16) where Frame i is known from input if i = 1, else it is known from previous step. The two axes at hand could be intersecting, parallel or skewed. Dual Vector Algebra, as explained in Section 2, is used to determine their relationship. Possible cases are explained below: a) Intersecting Axes: The origin ( ) of Frame i+1 is the point of intersection of lines and as explained in Section 2.2. The vector is the common normal to and and one of the two possible expression is given below: = (17) where is the unit vector along X i+1

6 b) Parallel Axes: A line ( ) is determined which is a common normal to both the lines and, as explained in Section 2.3. The origin ( ) of Frame 2 is then determined as the point of intersection of lines and. The vector, a unit vector from O i to O i+1, is then given as: = (18) c) Skewed Axes: The common normal of two skewed lines ( and ) is determined using the methodology of Section 2.4. The point of its intersection with is and that with is the origin ( ) of Frame i+1. The vector is then given as: = (19) Once the origin ( ) and vector are found, the Frame i+1 attached to Link i is completely known by determining as: = (20) where is a unit vector along Y i+1. Now both the Frames i and i+1 are known. The DH parameters associated with the transformation between them as illustrated in Figure 3(b), are given below: First, the joint offset b is found as b = ( ). (21) Next, the joint angle θ is obtained, i.e., θ =2(( ).,. ) (22) The link length a and twist angle are finally given by a = ( ). (23) = 2(( ).,. ) (24) The above procedure is repeated until the last joint (i = n). Note that for the last joint, the point O n+1 needs to be provided as input. For that, it is assumed that is parallel to [1, 2]. If the point O n+1 lies on last joint axis, a unique solution for cannot be found. In that case, Frame n+1 should be provided as input and its treatment is excluded here to keep the approach simple. The complete working of the proposed methodology as a generic algorithm is shown in Figure 4(a), which calls different functions, as shown in Figures 4(b) to 4(f). 4. IMPLEMENTATION IN AUTODESK INVENTOR The proposed analytical method has been implemented as an addin/plugin inside Autodesk Inventor software. The CAD assembly of a serial manipulator, as shown in Figure 5 was imported inside the Inventor environment. The base of the robot is placed such that the frame attached to it is identical to the world coordinate system. Assembly constraints are then defined between the links/parts of the manipulator in sequence from the base to the end-effector.

7 Start n = number of joints to, C 2 to C n, O n+1 Frame 1 = Frame F; O 1 = O F; = Frame 1 is attached to Link 0 i = 1; = ; = Angle(,, ) x =. y = ( ). angle = atan2(y, x) Return angle = [ ] T = [ ] T (c) Angle Between Vectors Function Frame + 1 = Revolute(, ) Intersecting(,,, ) b = ( ). θ = Angle(,, ) a = ( ). = Angle(,, ) Find from Section 2.2 = i = i + 1 Yes i < n? No End (d) Intersecting Joint Axes End (a) Main DH Extraction Loop Parallel(,,, ) Revolute(, ) Find from Section 2.3 is dual vector of is dual vector of = Line Dot Product Dual Part = 0? No End (e) Parallel Joint Axes No Intersecting(,,, ) Yes Real Part = 1? Yes Parallel(,,, ) = Frame 2= new frame with, and at Return Frame 2 (b) Revolute Joint Function Skew(,,, ) Skew(,,, ) Find from Section 2.4 = End (f) Skewed Joint Axes Figure 4: Algorithm for the extraction of DH parameters

8 The adddin developed using Visual C# C reads the C CAD assembly data using thee Application Programming P Interfacce (API) of Auttodesk Inventoor and lists the CAD parts and d the assemblyy constraints beetween them. In Autoodesk Inventor, Insert assem mbly constrainnt corresponds to a revolute jooint and is appplied between circles of equal radii located on two o mating links//parts. Using Innventor API, thhe properties of o each circle and hennce the associaated revolute jooint can be extrracted into the addin. The norrmal of each ciircle as a unit vector and a its center, both in the world w coordinaate system, aree determined. T These are then n provided as input to o the novel algo orithm and the DH parameterrs are determinned as explaineed in Section 3.. Worldd Coordinate Syystem Fig gure 5: Determ mination of circcles correspondding to revolutte (insert) joints in Autodesk Inventor Figurre 6: DH param meters extractioon from the CA AD model of K KUKA KR5 a consstraints of the CAD model of KUKA KR R5 manipulator is listed in The linnks/parts and assembly Figure 6 (callout 1). Its I exact DH parameters p as eextracted by th he addin, is shoown in Figure 6 (callout 2).

9 The transformation of frame attached to the end-effector with respect to the base achieved by multiplying the transformation matrices corresponding to each DH parameter set (callout 3 of Figure 6) matches exactly with the transformation matrix determined directly between these two frames using Inventor API (callout 4 of Figure 6). The coordinate frames determined were inserted into the CAD model as shown in Figure 6 (callout 5) which help in better visualization and understanding of the DH parameters. The DH parameters also match with those mentioned in the company s specifications of KUKA KR5 [8]. Similarly, the DH parameters of PUMA 500 industrial robot/manipulator was extracted and verified with its specifications available in literature [9]. 5. CONCLUSIONS In this paper, a novel analytical method to extract the Denavit-Hartenberg (DH) parameters is proposed. The method requires the joint axes and a point on it as input and uses the concept of Plücker coordinates to represent the joint axes of a serial manipulator. Dual Vector Algebra was then used to determine the relationship between two successive joint axes and then the DH parameters were determined using Vector Algebra, in a forward recursion from base link to the end-effector. The methodology was used as an addin developed for Autodesk Inventor, which extracts DH parameters from CAD assembly of industrial manipulators/robots. The DH parameters of KUKA KR5 and PUMA 500 manipulators were determined from their CAD models and were validated with the existing results. The methodology can be practically used to determine the exact DH parameters an actual manipulator by tracing a point on the end-effector while rotating one joint at a time. The exact parameters determined can then be used to make modifications in its controller or be used as nominal parameters for the error estimation during calibration. ACKNOWLEDGEMENT The research work reported here was carried out under a project sponsored by BARC/BRNS, India. The authors sincerely acknowledge their financial support. REFERENCES [1] Abderrahim, M.; Whittaker, A. R.: Kinematic Model Identification of Industrial Manipulators. Robotics and Computer Integrated Manufacturing, Vol. 16, pp. 1-8, [2] Barker, L. K.: Vector-Algebra Approach to Extract Denavit-Hartenberg Parameters of Assembled Robot Arms. Technical Paper #2191, NASA, [3] Bauchau, O.A.: Flexible Multibody Dynamics. Dodrecht: Springer, [4] Denavit, J; Hartenberg, R. S.: A Kinematic Notation for Lower-pair Mechanisms Based on Matrices. ASME Journal of Applied Mechanisms, Vol. 22, No. 2, pp , [5] Fischer, I.S.: Dual-Number Methods in Kinematics, Statics and Dynamics, Boca Raton, CRC Press, [6] Hayati, S.; Mirmirani M.: Improving absolute accuracy of robot manipulators. International Journal on Robotic Systems, Vol 2(4), pp , [7] Ketchel, J. S.; Larochelle, P. M.: Collision Detection of Cylindrical Rigid Bodies Using Line Geometry. Proceedings of ASME International Design Engineering Technical Conferences & Computers and Information in Engineering Conference (IDETC/CIE), pp. 1-15, [8] KUKA Robotics website, [9] PUMA webpage, [10] Saha, S.K.; Introduction to Robotics. New Delhi: Tata McGraw-Hill, 2008 [11] Stone, H. W.; Sanderson, A. C.; Neuman, C. P.: Arm Signature Identification. IEEE International Conference on Robotics and Automation, pp , 1986.